The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts seed-redirects.mjs and delivery-lint.mjs dynamically import eds-path.mjs by resolving its location at runtime relative to the current script's path. This technique for loading local dependencies introduces a risk as it executes code from computed paths.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from migrated HTML files and external source URLs through scripts like inventory.mjs, verify.mjs, and optimize.mjs. As these files are processed and their contents are potentially surfaced to the agent, there is a risk of indirect prompt injection if the ingested data contains malicious instructions. Evidence chain:
Ingestion points: stardust/migrated/*.html and external source URLs analyzed in section-fidelity.mjs and media-reconcile.mjs.
Boundary markers: None identified; files are processed directly.
Capability inventory: File system modifications via autofix-aem.mjs, network fetches via fetch, and script execution for deployment tasks.
Sanitization: Minimal sanitization (e.g., stripTags in limited cases) is applied before data is processed.
[EXTERNAL_DOWNLOADS]: The skill documentation explicitly requires the installation of external tools and other agent skills from non-trusted GitHub repositories, specifically github.com/pbakaus/impeccable and repositories under coreyhaines31/marketingskills.
[COMMAND_EXECUTION]: The procedure involves the execution of numerous shell commands and local scripts that perform file system modifications and network requests to external origins for site verification and image reconciliation.
[DATA_EXPOSURE]: While no secrets are hardcoded, the skill relies on the presence of a DA_TOKEN environment variable for interacting with AEM Edge Delivery Services APIs, which the agent is instructed to verify during setup.

rollout